Modeling of Clouds from a Single Photograph

In this paper, we propose a simple method for modeling clouds from a single photograph. Our method can synthesize three types of clouds: cirrus, altocumulus, and cumulus. We use three different representations for each type of cloud: two-dimensional texture for cirrus, implicit functions (metaballs) for altocumulus, and volume data for cumulus. Our method initially computes the intensity and the opacity of clouds for each pixel from an input photograph, stored as a cloud image. For cirrus, the cloud image is the output two-dimensional texture. For each of the other two types of cloud, three-dimensional density distributions are generated by referring to the cloud image. Since the method is very simple, the computational cost is low. Our method can generate, within several seconds, realistic clouds that are similar to those in the photograph.